摘要
二维可分离小波在纹理分析领域得到了成功的应用,但它只提取图像水平、垂直和对角方向的频率信息,其变换滤波器是各向同性的,不能很好地表达纹理的细节。利用剪切波变换优良的多尺度性、局域性和方向性,提出一种基于剪切波变换(S hearlet transform)的纹理分类算法。该方法先对纹理图像做剪切波变换,得到各尺度、方向子带的剪切系数,计算尺度间子带能量比,以尺度间能量比为权对各子带能量加权,以加权后的子带能量作为特征矢量,用K邻近分类器进行分类。实验结果表明该方法比基于小波的纹理分类方法更加有效。
Two-dimensional separable wavelets have obtained successful application in texture analysis,but they extract infor- mation only in three directions, along horizontal, vertical and diagonal, so they are not very good expression for texture details.This paper proposes a method based on Discrete Shearlet Transform（DST） for texture classification.Frequency coefficients can be got at all scales and in all directions after performing DST to original image.Energy contrast is calculated from different scales and energy is weighted according to the energy contrast.Sub-band energy weighted is taken as the characteristic vector and KNN classifier is used to classify textures.The tests show that the Shearlet-based algorithm is significantly more effective than the one wavelet-based.
出处
《计算机工程与应用》
CSCD
北大核心
2011年第29期15-17,68,共4页
Computer Engineering and Applications
基金
国家自然科学基金(the National Natural Science Foundation of China under Grant No.61003120)
关键词
剪切波变换
纹理分类
特征提取
Shearlet transform
texture classification
feature extraction
